Early Detection of Driver Drowsiness Utilizing Machine Learning based on Physiological Signals, Behavioral Measures, and Driving Performance

Abstract

Driver drowsiness is one of the most causes of traffic accidents worldwide. To prevent such accidents, it is necessary to detect driver drowsiness as early as possible. In previous studies, it was confirmed that decreasing arousal levels affect physiological indices, behavioral indices, and driving performance. The goal of this study is to classify the alert states of drivers, particularly the slightly drowsy state, based on physiological indices, behavioral measures, and driving performance. First, the relationship between the arousal level of a driver, physiological signals, such as electroencephalogram and electrocardiogram signals, behavioral measures, and driving performance is investigated based on analysis of data measured by a driving simulator (DS) and driver monitoring system. Next, to classify the alert and the slightly drowsy states utilizing machine learning algorithms, a total of 32 features are extracted from the measured data over a period of 10 seconds. Four machine learning algorithms, namely logistic regression, support vector machines, the k-nearest neighbor classifier, and random forest (RF), are utilized for the classification of driver drowsiness in this study. As a result, it is confirmed that the RF method can obtain up to 81.4% accuracy when distinguishing between alert and slightly drowsy states. This result demonstrates the feasibility of driver drowsiness detection based on hybrid measures over a 10-second time period with high accuracy.